yenklabs/dali-verification-taxonomy

Part of the Dali Open Evidence Ecosystem

Dali is the open verification layer for AI: it creates, scores, and preserves evidence so AI-assisted outputs can be independently verified, exchanged, and replayed. Legal AI is the proving ground.

Dali Verification Taxonomy

Five verification outcomes used by Dali to classify whether evidence behind an AI output can be trusted.

Applies beyond legal citations: any domain where AI outputs must be verified, exchanged, and preserved.

Evidence failures are not binary. Link checkers return valid or broken. Verification requires classification that separates what broke from why it broke.

Purpose

The Dali Verification Taxonomy provides a standardized language for classifying verification outcomes during AI-assisted legal review.

It enables consistent reporting, reproducible evaluation, and comparable verification results across tools and organizations.

Outcomes

ID	Label	Detection stage
verified	Verified	existence_and_support
authority_not_found	Authority Not Found	existence
proposition_unsupported	Proposition Unsupported	support
source_trail_missing	Source Trail Missing	evidence_preservation
unverifiable	Unverifiable	evidence_preservation

Files

• outcomes.jsonl — one row per outcome for programmatic use (Dataset Viewer)
• meta/taxonomy.json — full taxonomy document

Engineering note

• Outcomes 2 and 3 require different detection pipelines
• Outcomes 4 and 5 are evidence preservation failures, not model failures
• Outcome 1 still requires a sealed bundle — correctness without provenance is not defensible

Intended Users

This project is designed for:

• Legal AI researchers
• AI evaluation platforms
• Law firms
• Legal technology companies
• Academic researchers
• Responsible AI practitioners
• Courts and public-interest organizations
• Open-source contributors

The Dali Open Evidence Ecosystem

Each project serves a different role while remaining independently useful.

Resource	Purpose	Status
Dali Open Evidence Corpus	Preserve reproducible evidence artifacts	available
Dali Citation Benchmark (seed sample)	Hand-curated public sample for methodology review (5 cases, 14 authorities)	available
Dali Verification Taxonomy	Standardize verification outcomes	available
Full evaluation run	524 citations · 3 models · 5 jurisdiction tracks — data/results/	available
Dali Platform	Open-source verification engine	available (early-stage v0.x)
Dali Evaluation Prompts	Cross-jurisdiction evaluation resources	planned
Dali Replay Corpus	Replay traces and policy-version hashes	planned
Dali Evidence Artifacts	Portable evidence packages	planned
Research models (taxonomy, citation risk, authority matching, proposition support)	Lightweight research classifiers	planned

GitHub

https://github.com/yenklabs/Dali

Website

https://yenklabs.com

How Everything Fits Together

AI-Assisted Work
            │
            ▼
Seed Benchmark Sample / Full Evaluation Run
            │
            ▼
Verification Taxonomy
            │
            ▼
Open Evidence Corpus
            │
            ▼
Evidence Ledger
            │
            ▼
Evidence Infrastructure (Dali)

The seed sample is for methodology review.

The full evaluation run is the measured result set.

The taxonomy classifies.

The corpus preserves evidence.

Together they form Dali's open evidence infrastructure.

Current Project

• Open-source project (early-stage v0.x · current release v0.2.1)
• Public seed corpus and full evaluation run (named separately)
• Cross-jurisdiction support
• Open verification taxonomy
• Open evidence corpus
• CC BY 4.0 licensed datasets

Citation

If you use this dataset in research, benchmarking, or publications, please cite:

@dataset{dali_verification_taxonomy_v0_1,
  title        = {Dali Verification Taxonomy},
  author       = {{Dali Contributors}},
  organization = {GammaLex AI Inc.},
  year         = {2026},
  version      = {0.1},
  url          = {https://huggingface.co/datasets/yenklabs/dali-verification-taxonomy},
  note         = {Open taxonomy for classifying AI verification outcomes}
}

Dataset version 0.1 is the seed corpus release. The Dali engine release is v0.2.1 (github.com/yenklabs/Dali).

Version History

Version	Description
v0.1	Initial public seed release
v0.2 / v0.2.1	Evaluation harness and documentation improvements (current)
Later v0.x	Continued early-stage releases — not a 1.0 claim

Roadmap

Available

• ✓ Open Evidence Corpus (seed)
• ✓ Citation Benchmark seed sample (5 cases / 14 authorities)
• ✓ Full evaluation run (524 citations · data/results/)
• ✓ Verification Taxonomy
• ✓ Open-source verification platform (v0.2.1)

Planned

• Proposition verification
• Cross-jurisdiction expansion
• Community-contributed evidence
• Quarterly benchmark reports
• Evidence Ledger
• Evidence replay and reconstructability
• Dali Evaluation Prompts
• Dali Replay Corpus
• Dali Evidence Artifacts
• Research models (taxonomy classifier, citation risk, authority matching, proposition support)

Contributing

We welcome community contributions.

Areas where contributors can help include:

• Additional jurisdictions
• Verification methodologies
• Evidence artifacts
• Benchmark prompts
• Documentation improvements
• Dataset validation
• Research collaboration

GitHub

https://github.com/yenklabs/Dali

Design Principles

The Dali ecosystem is built around:

• Reproducibility
• Transparency
• Independent verification
• Evidence preservation
• Human accountability
• Cross-jurisdiction support
• Open collaboration

Why Dali Exists

AI-assisted work increasingly requires more than model outputs.

Organizations need to understand:

• What the AI generated.
• What evidence was reviewed.
• What could be independently verified.
• What could not be verified.
• What changed during review.
• What humans ultimately decided.

Dali is the open verification layer for AI: it creates, scores, and preserves evidence so AI-assisted outputs can be independently verified, exchanged, and replayed.

Related resources

• Essay: yenklabs.com/notes/dali-five-verification-outcomes
• Open evidence corpus: huggingface.co/datasets/yenklabs/open-evidence-corpus
• Benchmark: yenklabs/dali-citation-benchmark
• EPS draft: yenklabs.com/artifacts/evidence-package-spec-v0.1
• Ecosystem: huggingface.co/yenklabs